Best available HA solution with 3 servers

[andrea_it]

Hello,
I have some doubts on how High Availability can work in ProxMox.
I'd like to create an HA enviroment to host my personal websites and also to lear more about HA.
I imagine I have 3 identical server and all of them are part of the same cluster.
I read about HA with local storage, shared storage, hyperconverged setup so I'm a bit confused.
I already read the manual, but I think I'm still missing some piece so I'm asking you to help me understand.

What's in unclear to me is the following:

1 - How are the vm images stored? Are the vm images replicated to each node and kept in sync?
2 - When I create a new VM with HA enabled do I have to specify the underlying node or it's automatically selected?
3 - Is this vm copied/synced over all the nodes immediately?
4 - Can the vm be moved to another node if performance becomes "unacceptable"? Can we move VM based on kinda of QoS rules or HA is for failure only?

Thanks in advance for your help

 

[guletz]

1 - How are the vm images stored? Are the vm images replicated to each node and kept in sync?

As a block device or as a file(dependig what kind of storge do you want to use)
Depending on what do you use(local storage, like zfs) you can setup that ANY of your VM to be replicated(async, time base schedule) from node A to node B

 

[UdoB]

Disclaimer: I am just a normale user at home, no commercial setting, no public services except for family.

1 - How are the vm images stored? Are the vm images replicated to each node and kept in sync?

You have several choices, the most obvious:

• use ZFS as the underlying Filesystem and use the Proxmox replication feature to replicate every few hours or every 15 minutes. I do replicate only once a day. On Failover/Migration the missing parts gets transferred...
• use a Network base storage like FreeNAS for NFS or iSCSI
• use the very clever Proxmox Ceph system directly on your nodes --> Hyper Converged

You need to read more about those options to know which one is the "right" one for you!

2 - When I create a new VM with HA enabled do I have to specify the underlying node or it's automatically selected?

Usually I know which Node I want to put a new HA-machine on. When this node fails HA will move it to one of the other two nodes. Yes, there must be some logic to decide which one to choose but I don't know.

3 - Is this vm copied/synced over all the nodes immediately?

Depends on the technology:

• ZFS with local storage only: No! It syncs in discrete intervals. The interval will be configured by you
• a "real" network filesystem (Gluster, Ceph, NFS, ...) would store new data on all disk immediately (Simplified assumption - this depends on several factors and distribution rules)

4 - Can the vm be moved to another node if performance becomes "unacceptable"? Can we move VM based on kinda of QoS rules or HA is for failure only?

 

[guletz]

2 - When I create a new VM with HA enabled do I have to specify the underlying node or it's automatically selected?

First you must create the VM, so you need to specify on what node do you want to host this VM. Then you most configure the HA part(and of course the replication if you use zfs for this VM, or not if you use a shared storage like NFS)

 

[spirit]

>>1 - How are the vm images stored? Are the vm images replicated to each node and kept in sync?

you need a shared storage. (could be a san, or you can use ceph on your proxmox nodes to replicate local disk)

>>2 - When I create a new VM with HA enabled do I have to specify the underlying node or it's automatically selected?
it'll be randomly choose, but you can create groups of servers, and assign this vm in a group

>>3 - Is this vm copied/synced over all the nodes immediately?
if you use ceph, yes. (if you use zfs replication, it's async,so you can loose datas)

>>4 - Can the vm be moved to another node if performance becomes "unacceptable"? Can we move VM based on kinda of QoS rules or HA is for >>failure only?

no, currently they are no "loadbalancing" feature.
it's only failure only for now.